CN108335137B

CN108335137B - Sorting method and device, electronic equipment and computer readable medium

Info

Publication number: CN108335137B
Application number: CN201810097006.3A
Authority: CN
Inventors: 唐金川
Original assignee: Beijing Sankuai Online Technology Co Ltd
Current assignee: Beijing Sankuai Online Technology Co Ltd
Priority date: 2018-01-31
Filing date: 2018-01-31
Publication date: 2021-07-30
Anticipated expiration: 2038-01-31
Also published as: WO2019148982A1; CN108335137A

Abstract

The disclosure relates to a sorting method and apparatus, an electronic device, and a computer-readable medium. Belonging to the technical field of Internet, the sequencing method comprises the following steps: acquiring a plurality of objects to be sequenced according to the request information; acquiring a sorting characteristic value corresponding to each object; and sorting the plurality of objects according to the sorting feature values; the sorting characteristic value corresponding to each object is obtained based on the click characteristic value and the corresponding click weight of each object, the conversion characteristic value and the corresponding conversion weight, and the transaction characteristic value and the corresponding transaction weight. The method and the device can sort the objects to be sorted based on the transaction characteristic values, so that the sorting accuracy is further improved.

Description

Sorting method and device, electronic equipment and computer readable medium

Technical Field

The present disclosure relates to the field of internet technologies, and in particular, to a sorting method and apparatus, an electronic device, and a computer-readable medium.

Background

The sequencing optimization mechanism is a particularly important ring in a searching and recommending system, particularly an advertising system. For effective advertising, most of the existing advertising platforms are charged according to the Click-through (CPC (Cost Per Click) settlement mode, and the advertisement is charged according to the number of times of being clicked. In the existing effect advertisement system, according to platform benefits and user experience, the following two sorting mechanisms are mainly used at present:

1. and (4) according to a click rate optimization sorting mechanism.

Search engines such as google, hundredths, whose user experience is mainly measured by click-through rate. It is a better experience for users to find the web page they want for search faster and click to leave quickly. Therefore, the advertisement platforms generally select to optimize the click rate of the whole platform, the click rate is increased, the whole user experience is optimized to a certain extent, and the platform can obtain more advertisement income. The basic sorting formula of such a sorting mechanism is:

rank score is bid click-through rate.

2. And (4) a sorting mechanism optimized according to click quantity and conversion quantity.

For the platform capable of tracking the transaction, such as amazon and Taobao electronic commerce websites, for the measurement of the user experience, in addition to the click rate measurement in the above sort of sorting mechanism, the index of the conversion rate is additionally added as the measurement of the user experience. Briefly, the user experience is mainly the experience of "shopping" and "buying", so when optimizing the user experience, such platforms select to optimize the whole click quantity and the conversion quantity, and combine the click quantity and the conversion quantity in a weighted summation mode through different weights w1 and w 2. Meanwhile, the optimization of the conversion amount can also give more conversion units to the advertiser, thereby bringing more profits to the advertiser to a certain extent.

User experience score w1 click-through + w2 conversion

A basic formula for such a sorting mechanism is:

rank score bid (w1 click through + w2 conversion)

The weight of the sorted scores w1 and w2 is also critical, which affects the importance of the click and conversion in the user experience as a whole, and these two values are usually manually counted and manually set according to the target setting. The two values w1 and w2 are usually analyzed by an algorithm person according to data statistics, and then specific values are manually set according to the target to be optimized (for example, w1 is 0.3, and w2 is 0.7).

Disclosure of Invention

The present disclosure provides a sorting method and apparatus, an electronic device, and a computer-readable medium, which can at least partially or completely solve the above problems in the prior art.

Additional features and advantages of the disclosure will be set forth in the detailed description which follows, or in part will be obvious from the description, or may be learned by practice of the disclosure.

According to an aspect of the present disclosure, there is provided a sorting method including: acquiring a plurality of objects to be sequenced according to the request information; acquiring a sorting characteristic value corresponding to each object; and sorting the plurality of objects according to the sorting feature values; the sorting characteristic value corresponding to each object is obtained based on the click characteristic value and the corresponding click weight of each object, the conversion characteristic value and the corresponding conversion weight, and the transaction characteristic value and the corresponding transaction weight.

In an exemplary embodiment of the present disclosure, the obtaining the ranking characteristic value corresponding to each object includes: acquiring historical exposure, historical click rate and historical lower unit rate of each object; and obtaining a click characteristic value, a conversion characteristic value and a transaction characteristic value of each object according to the request information, the historical exposure, the historical click quantity and the historical order placing quantity.

In an exemplary embodiment of the present disclosure, obtaining the click feature value of each object according to the request information, the historical exposure amount, and the historical click amount includes: acquiring the historical exposure click rate of each object according to the historical exposure and the historical click rate; and obtaining the click characteristic value of each object according to the object attribute of each object, the historical exposure click rate and the request information.

In an exemplary embodiment of the present disclosure, obtaining the conversion feature value of each object according to the request information, the historical exposure amount, the historical click rate, and the historical order quantity comprises: obtaining the historical click order rate of each object according to the historical click amount and the historical order rate; obtaining the historical exposure conversion rate of each object according to the historical exposure click rate and the historical click order rate; and obtaining a conversion characteristic value of each object according to the object attribute of each object, the historical exposure conversion rate and the request information.

In an exemplary embodiment of the present disclosure, obtaining the transaction characteristic value of each object according to the request information, the historical exposure amount, the historical click rate, and the historical order quantity comprises: obtaining a predicted transaction amount of each object according to the object attribute of each object and the request information; normalizing the predicted transaction amount for each subject; and obtaining the transaction characteristic value according to the normalized predicted transaction amount and the conversion characteristic value of each object.

In an exemplary embodiment of the present disclosure, further comprising: and acquiring the click weight, the conversion weight and the transaction weight of each object according to the current state of each object.

In an exemplary embodiment of the present disclosure, the obtaining of the click weight, the conversion weight, and the transaction weight of each object according to the current state of each object includes: when the object is in a first state, setting the click weight of the object to be greater than the conversion weight and the conversion weight to be greater than the transaction weight; and/or when the object is in the second state, setting the conversion weight of the object to be more than or equal to the transaction weight and the transaction weight to be more than the click weight; and/or when the object is in a third state, setting the transaction weight of the object to be greater than the conversion weight and the conversion weight to be greater than the click weight; and the sum of the click weight, the conversion weight and the transaction weight of each object is a preset constant.

In an exemplary embodiment of the present disclosure, when the object is in the first state, setting the click weight of the object to be greater than the conversion weight and the conversion weight to be greater than the transaction weight includes: and when the consumption budget ratio of the object is in a first preset range, increasing the click weight of the object according to the consumption budget ratio of the object.

In an exemplary embodiment of the present disclosure, when the object is in the second state, setting the conversion weight of the object to be equal to or greater than the transaction weight and the transaction weight to be greater than the click weight includes: and when the consumption budget ratio of the object is in a second preset range, increasing the conversion weight of the object according to the conversion characteristic value of the object.

In an exemplary embodiment of the present disclosure, when the object is in the third state, setting the transaction weight of the object to be greater than the conversion weight and the conversion weight to be greater than the click weight includes: and when the consumption budget ratio of the object is in a third preset range, increasing the transaction weight of the object according to the input-output ratio of the object.

In an exemplary embodiment of the present disclosure, the request information includes search information input by a current user and/or combination information between the current user and each object and/or a user attribute of the current user.

According to an aspect of the present disclosure, there is provided a sorting apparatus including: the object acquisition module is used for acquiring a plurality of objects to be sequenced according to the request information; the characteristic value acquisition module is used for acquiring a sequencing characteristic value corresponding to each object; the sorting module is used for sorting the plurality of objects according to the sorting characteristic value; the sorting characteristic value corresponding to each object is obtained based on the click characteristic value and the corresponding click weight of each object, the conversion characteristic value and the corresponding conversion weight, and the transaction characteristic value and the corresponding transaction weight.

According to an aspect of the present disclosure, there is provided an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the program implementing the method steps of any of the above embodiments when executed by the processor.

According to an aspect of the present disclosure, a computer-readable medium is provided, on which a computer program is stored, which program, when being executed by a processor, is adapted to carry out the method steps of any of the above embodiments.

According to the sorting method and device, the electronic equipment and the computer readable medium in some embodiments of the disclosure, the sorting characteristic value of each object is obtained based on the transaction characteristic value of the object to be sorted and the corresponding transaction weight, so that on one hand, more accurate object sorting can be realized; on the other hand, more accurate putting in can realize promoting.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure.

Drawings

The above and other objects, features and advantages of the present disclosure will become more apparent by describing in detail exemplary embodiments thereof with reference to the attached drawings.

FIG. 1 is a flow chart illustrating a method of sorting according to an exemplary embodiment.

FIG. 2 is a flow chart illustrating another method of ordering according to an exemplary embodiment.

FIG. 3 is a schematic diagram illustrating a sequencing apparatus according to an exemplary embodiment.

FIG. 4 is a schematic diagram of an electronic device shown in accordance with an exemplary embodiment.

Detailed Description

Example embodiments will now be described more fully with reference to the accompanying drawings. Example embodiments may, however, be embodied in many different forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the concept of example embodiments to those skilled in the art. The drawings are merely schematic illustrations of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and thus their repetitive description will be omitted.

Furthermore, the described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a thorough understanding of embodiments of the disclosure. One skilled in the relevant art will recognize, however, that the subject matter of the present disclosure can be practiced without one or more of the specific details, or with other methods, components, devices, steps, and the like. In other instances, well-known structures, methods, devices, implementations, or operations are not shown or described in detail to avoid obscuring aspects of the disclosure.

In the design of the existing sequencing mechanism, besides optimizing platform benefits, optimization is mainly carried out from the perspective of user experience. The platform which can not track the transaction mainly optimizes the click quantity, and the platform which can track the transaction additionally optimizes the conversion quantity. While in the three parties (user, advertiser, platform) that participate in the advertising marketing campaign, additional objectives of the advertiser (e.g., trade volume, ROI) are not considered in the ranking optimization.

The foregoing ranking mechanism that optimizes user conversions is not necessarily optimal for the advertiser. For example, for an advertiser, if the gross margin for a 10 conversion list is not as high as the gross margin for a 1 list, the advertiser would prefer to choose to promote the 1 conversion list as much as possible. Therefore, the conversion amount is optimized, and the optimization target of the advertiser cannot be completely met.

Only if the three parties involved in the advertising marketing campaign benefit better is a win-win situation, allowing the marketing campaign to continue to develop.

In a second prior art ranking mechanism, each advertiser has a ranking score (bid (w1 click through + w2 conversion), and then ranks the ranking scores in descending order according to the ranking score.

For example, a merchant rank of 0.7 for a and 0.6 for b, then a is ranked before b. This ranking score is calculated according to the "ranking score" formula described above. The ranking score formula is also calculated according to the click rate, conversion rate and bid combination.

For another example, if the bid of a merchant is 1 (which may be 1-tuple/click per time), the click rate is 0.8, and the lower rate is 0.6, the rank of a is 1 × (w1 × 0.8+ w2 × 0.6); assuming that the bid, click rate, and lower rate of b merchants are 1, 0.6, and 0.6, respectively, the rank of b is 1 (0.6 w1+0.6 w 2). In the second existing sorting mode, w1 and w2 have the same value for all merchants. Namely, the above prior art solutions have at least the following disadvantages:

1. the prior art scheme is mainly based on the user experience, and the appeal of an advertiser on transaction amount optimization is not added to the optimization of a sequencing mechanism.

2. Advertisers are in different stages of development, and the appeal to different targets is different, and the existing scheme does not consider the appeal. In the second current sorting mechanism, the settings of w1 and w2 are automatically adjusted over time, but the values of w1 and w2 are the same for all merchants. This makes it impossible to optimize the personalized appeal of the advertiser with different click rate and conversion rate.

As shown in fig. 1, the sorting method may include the following steps.

In step S110, a plurality of objects to be sorted is acquired according to the request information.

Internet websites typically present a plurality of objects to a user to enable the user to browse and perform corresponding conversion operations. For example, in an e-commerce website, the object may be a product recommended to the user, and the user who logs in the e-commerce website may perform a conversion operation such as further purchasing by browsing related information of the product. For example, an Application (APP) installed on a mobile terminal (e.g., a mobile phone, a tablet computer, a wearable smart device, etc.) may recommend a corresponding store or shop, such as a restaurant, to a user based on a current geographic location of the user or a search keyword input by the user.

At present, when a plurality of objects are displayed on each website or APP, the objects are often based on a certain ranking rule, for example, after a user searches in a search engine, a search result is displayed according to a preset ranking mode.

In this embodiment, a method for sorting a plurality of objects is actually used, and therefore, before sorting, a plurality of objects to be sorted are obtained. For example, all objects under the same category are acquired as a plurality of business objects to be sorted.

In the present embodiment, the object refers to various objects that can be presented to the user through the internet and perform a corresponding conversion operation by the user, and may be, for example, a product, an application, a store, a service, and the like presented to the user.

In an exemplary embodiment, the request information includes search information input by a current user and/or combination information between the current user and each object and/or a user attribute of the current user.

Specifically, the search information may be a search keyword currently input by the user, such as words or sentences like "hot pot", "cake", "fresh flower", "mobile business hall around the tokyo". It should be noted that the search keyword herein is not necessarily input by the user currently, but may also be a search keyword that is input by the user historically, or a combination of the currently input keyword by the user and the historically searched keyword, or a search keyword within a historical period, such as a keyword input last time or a keyword input last week.

The combined information may be information of a distance, a direction, a traffic condition, and the like between the current user and each object, and may further include a matching degree or a correlation between a search keyword input by the user and each object, for example, information of whether the user likes a category corresponding to each object, whether the user likes consumption in a business circle where the object is located, and the like.

In an exemplary embodiment, the user attributes include any one or more of taste preferences, environmental preferences, price sensitivity, brand preferences, etc. of the current user.

In this embodiment, the user attribute may include personalized information of the current user, such as taste preference (which may be analyzed statistically according to information such as historical purchase records and ordering records of the current user, for example, the user prefers spicy chinese cuisine), environment preference (for example, some users pay more attention to shopping or eating environment and want to consume in quiet shops), price sensitivity (for example, some users may pay less attention to eating environment and pay more attention to cost performance, while other users may not be sensitive to price), brand preference (for example, users may pay more attention to clothes of a certain brand under the same category of clothes under the same condition), category preference (for example, the user prefers a robust cuisine), business circle preference (for example, the user is currently in the middle of two business circles, but the user prefers one of the business circles), Distance sensitivity (for example, some users pay attention to traffic convenience, and some users can eat food of the heart instrument no matter how far away from the heart instrument), and the like.

And acquiring a plurality of objects to be sequenced according to the request information, wherein a plurality of application scenes exist. For example, a current user may open a certain APP on a mobile phone, and the multiple objects to be ranked may be obtained on a home page of the APP according to user attributes of the current user (e.g., taste preferences of the user) and combination information between the user and the corresponding objects (e.g., distance between the user and a store around the user). For another example, a current user may open a certain APP on a mobile phone, and the multiple objects to be sorted may be obtained on a home page of the APP according to an operation record of the current user, such as a record of placing an order, in the last week or last time. For another example, only stores matching the keyword may be obtained as the plurality of objects to be ranked according to the search keyword input by the current user. For another example, the plurality of objects to be ranked may be obtained by comprehensively considering search information of the current user, combination information between the current user and each object, and the user attribute of the current user. The present disclosure is not limited thereto.

In step S120, a ranking feature value corresponding to each object is obtained.

The sorting characteristic value corresponding to each object is obtained based on the click characteristic value and the corresponding click weight of each object, the conversion characteristic value and the corresponding conversion weight, and the transaction characteristic value and the corresponding transaction weight.

In an exemplary embodiment, the obtaining the ranking characteristic value corresponding to each object includes: acquiring historical exposure, historical click rate and historical order quantity of each object for a current user; and obtaining a click characteristic value, a conversion characteristic value and a transaction characteristic value of each object according to the request information, the historical exposure, the historical click quantity and the historical order placing quantity.

In an exemplary embodiment, obtaining the click feature value of each object according to the request information, the historical exposure amount, and the historical click amount includes: acquiring the historical exposure click rate of each object according to the historical exposure and the historical click rate; and obtaining the click characteristic value of each object according to the object attribute of each object, the historical exposure click rate and the request information.

In this embodiment, the click feature value of each object may be obtained by using the first machine model. The first machine model is a model which is trained by corresponding machine learning algorithm or deep learning algorithm by adopting training data in advance, and any one of the machine learning algorithm or the deep learning algorithm can be adopted, which is not limited by the disclosure.

In this embodiment, the object attribute may include information such as a category, a business district, and a geographic location of each object.

It should be noted that the specific descriptions of the request information, the search information, the combination information, the user attribute, and the object attribute are all used for illustration, and may be adjusted and set according to specific situations in different application scenarios.

In an exemplary embodiment, obtaining the conversion feature value of each object according to the request information, the historical exposure amount, the historical click rate, and the historical order quantity comprises: obtaining the historical click order rate of each object according to the historical click amount and the historical order rate; obtaining the historical exposure conversion rate of each object according to the historical exposure click rate and the historical click order rate; and obtaining a conversion characteristic value of each object according to the object attribute of each object, the historical exposure conversion rate and the request information.

In this embodiment, the second machine model may be used to obtain the transformation feature value of each object. The second machine model is a model which is trained by corresponding machine learning algorithm or deep learning algorithm by adopting training data in advance, and any one of the machine learning algorithm or the deep learning algorithm can be adopted.

In an exemplary embodiment, obtaining the transaction characteristic value of each object according to the request information, the historical exposure amount, the historical click rate, and the historical order placement amount includes: obtaining a predicted transaction amount of each object according to the object attribute of each object and the request information; normalizing the predicted transaction amount for each subject; and obtaining the transaction characteristic value according to the normalized predicted transaction amount and the conversion characteristic value of each object.

In this embodiment, a third machine model may be used to obtain the predicted transaction amount for each object. The third machine model is a model which is trained in advance by adopting training data through a corresponding machine learning algorithm or a deep learning algorithm, and any one of the machine learning algorithm and the deep learning algorithm can be adopted, which is not limited by the disclosure.

It should be noted that the first machine model, the second machine model, and the third machine model may respectively adopt different machine learning algorithms or deep learning algorithms, or may also adopt the same machine learning algorithm or deep learning algorithm, which is not limited in this disclosure.

In an exemplary embodiment, further comprising: and acquiring the click weight, the conversion weight and the transaction weight of each object according to the current state of each object.

In an exemplary embodiment, the obtaining the click weight, the conversion weight, and the transaction weight of each object according to the current state of each object includes: when the object is in a first state, setting the click weight of the object to be greater than the conversion weight and the conversion weight to be greater than the transaction weight; when the object is in a second state, setting the conversion weight of the object to be more than or equal to the transaction weight and the transaction weight to be more than the click weight; when the object is in the third state, setting the transaction weight of the object to be greater than the conversion weight and the conversion weight to be greater than the click weight; and the sum of the click weight, the conversion weight and the transaction weight of each object is a preset constant.

For example, the preset constant may be 1, or may also be 2, or any other constant, which is not limited by the present disclosure. In the following embodiments, the preset constant is illustrated as 1.

In an exemplary embodiment, when the object is in the first state, setting the click weight of the object to be greater than the conversion weight and the conversion weight to be greater than the transaction weight includes: and when the consumption budget ratio of the object is in a first preset range, increasing the click weight of the object according to the consumption budget ratio of the object.

In an exemplary embodiment, when the object is in the second state, setting the conversion weight of the object to be equal to or greater than the transaction weight and the transaction weight to be greater than the click weight includes: and when the consumption budget ratio of the object is in a second preset range, increasing the conversion weight of the object according to the conversion characteristic value of the object.

In an exemplary embodiment, when the object is in the third state, setting the transaction weight of the object to be greater than the conversion weight and the conversion weight to be greater than the click weight includes: and when the consumption budget ratio of the object is in a third preset range, increasing the transaction weight of the object according to the input-output ratio of the object.

For example, when the budget consumption of an object is lower, the object is considered to be in the first state, and the advertiser is far from acquiring the attention/click traffic expected from the budget. When the budget consumption ratio of a certain object is moderate, the object can be considered to be in the second state, the click weight is less important at the moment, and the conversion weight is optimized. When the budget consumption ratio of a certain object is ultrahigh, the object can be considered to be in the third state, the ROI is low, and at this time, the transaction weight can be optimized in a focused manner, so that the transaction amount of the object is increased.

In step S130, the objects are sorted according to the sorting feature value.

In this embodiment, the method may further include: obtaining a ranking score of each object according to the ranking characteristic value and the bid of the corresponding object; each object is sorted in descending order according to the sorting score from large to small, but the present disclosure is not limited thereto.

In an exemplary embodiment, the method may further include: outputting the sorted objects to a client so that the sorted objects can be displayed on the client.

According to the ranking method provided by the embodiment, the click characteristic value and the corresponding click weight of each object, the conversion characteristic value and the corresponding conversion weight and transaction characteristic value and the corresponding transaction weight are obtained, and the corresponding ranking characteristic value is obtained based on the click characteristic value and the corresponding click weight of each object, the conversion characteristic value and the corresponding conversion weight and transaction characteristic value and the corresponding transaction weight of each object, so that the advertiser optimization target angle is used, the advertiser appeal for transaction amount optimization is increased to the ranking mechanism, the ranking result is more reasonable and accurate, and the advertisement promotion delivery accuracy is improved.

As shown in fig. 2, the sorting method may include the following steps.

In step S210, a plurality of objects to be sorted are obtained according to the request information.

In step S220, the historical exposure amount, the historical click amount, and the historical order amount of each subject are acquired.

In step S230, a click feature value, a conversion feature value, and a transaction feature value of each object are obtained according to the request information, the historical exposure amount, the historical click amount, and the historical order placement amount.

In step S240, the click weight, the conversion weight, and the transaction weight of each object are obtained according to the current state of each object.

In step S250, a ranking feature value corresponding to each object is obtained according to the click feature value and the corresponding click weight of each object, the conversion feature value and the corresponding conversion weight, and the transaction feature value and the corresponding transaction weight.

Here, the object is taken as an example of an advertisement delivered in a certain platform. When the advertisements are sorted, each advertisement needs to be assigned with a score and sorted by the score. The goal that the advertiser wants to optimize is added to the score design, and the optimization goal of the advertiser can be achieved.

For example, for the ith ad, the score at the time of ranking is calculated by the following formula:

rankScore_i＝bid_i×T_i

in the above formula, rank score_iRank score, bid, for the ith advertisement_iFor example, assuming 150 advertisement budgets per day for the ith advertisement, the bid is 1 click, and a maximum of 150 clicks per day, the budget of the day is consumed. T is_iIs the ranking characteristic value of the ith ad.

In this embodiment, the consumption is the portion of money that the system platform draws from the advertiser, and the daily (or weekly or monthly) drawing consumption is less than or equal to the daily (or weekly or monthly) advertising budget.

In this embodiment, the transaction amount (price) is normalized by a normalization factor (price normadivsor), and personalized click right of the advertiser (or object, store, advertisement) granularity is set according to the click amount, conversion amount, and transaction amount of the advertiser (or store), respectivelyHeavy alpha_iTransformation weight beta_iTransaction weight gamma_iWherein the CRT_i、CVR_i、price_iModel pre-evaluation values respectively representing three indexes of exposure click rate, click order rate and order transaction amount of ith advertisement and sorting characteristic value T of ith advertisement_iCan be calculated by the following formula:

in the above formula, the normalization factor priceNormDivisor is consistent for all advertisers or advertisements. Wherein the CRT_i、CVR_i、price_iThe probability values are respectively predicted by a machine learning model according to the historical exposure, the historical click rate, the historical unit amount, the historical conversion amount, the historical transaction amount and other data of the store corresponding to the ith advertisement. Wherein the CRT_iClick feature value, CTR, for that store_i×CVR_iFor the converted feature value of the store,

is the transaction characteristic value of the store.

Assuming that the historical exposure of the store corresponding to the ith advertisement is 100 times, the exposure means that the advertisement is displayed on a website or an APP page, and if there are 10 historical clicks, the exposure click rate of the store corresponding to the ith advertisement is 10/100-10%; if there are 1 historical order-placing amount in the 10 historical click amounts, the historical click order-placing rate of the store is 1/10-10%; the historical exposure conversion for that store is the exposure click rate multiplied by the click rate, 10% x 10%, 1%. CRT in the above formula_i、CVR_iThe probability value of the store is estimated by the machine model based on historical data (e.g., the above-mentioned historical exposure, historical click rate, historical order amount, historical exposure click rate, historical exposure conversion rate, etc.), the attributes of the store, the user attributes of the current user, and the combination information of the store and the current user. price_iThe single consumption amount of the current user is estimated through a machine model, and the single consumption amount of the current user can be estimated according to the historical consumption amount of the current user, the historical single consumption amount, the user attribute of the current user, the store attribute of the store, the combined information of the store and the current user, the matching degree of the search keyword input by the current user and the store, and the like.

In this embodiment, the ranking characteristic value T of the ith advertisement_iThe calculation formula is for a single store corresponding to a single request of the current user. Wherein, when the key word input by the user changes, the CRT_i、CVR_i、price_iChanges occur, for example, when the user inputs "hot pot" and "hot pot view" that the search result rankings given are not the same. In addition, when the position of the user when a search request is issued changes, the CRT_i、CVR_i、price_iChanges may also occur, such as CRT, which store is generally closer to the user_i、CVR_i、price_iThe greater the estimated probability value; and the farther away which store is from the user, the CRT_i、CVR_i、price_iThe smaller the estimated probability value. Meanwhile, even if the user inputs the same keyword at the same time and the same place, CRT is aimed at different users_i、CVR_i、price_iIt is also highly likely that the user attributes will be different for different users. That is, each time a search request is received, the system performs estimation by using a machine model or a deep learning method, and performs estimation by comprehensively considering store attributes, user attributes, and combination information of the store and the user, such as a distance, an orientation, a matching degree, and the like between the store and the user. The purpose of this is to achieve a match of the ranking results with the user requirements.

For example, for a specific store, historical evaluation, rating, star rating, comments, and the like of different users on the store can be used as the attributes of the store, and for a specific user, historical evaluation, rating, ordering, browsing, collecting, and the like of the user in different stores can be used for counting the taste preference of the user, for example, some users like hot pot, some users like Sichuan dish; user sensitivity to price, e.g., some users prefer to be cheap, environmental preferences, e.g., some users prefer a store with good environment, etc.

In this embodiment, the transaction amount is normalized to limit the transaction amount within a certain range after being processed. The first normalization is for the convenience of the subsequent data processing, and the criteria for different stores are uniform in the sorting. The normalization is not limited to one of the above equations, and may be performed by other suitable methods. For example, a linear function transformation, the expression is as follows:

y＝(x-MinValue)/(MaxValue-MinValue)

in the above formula, x and y are values before and after conversion, respectively, and MaxValue and MinValue are maximum and minimum values of the sample, respectively. For example, if the current user historical single consumption amount is between 10 and 200 yuan, MaxValue is 200, MinValue is 10, x is the current predicted single consumption amount, and y is the normalized current predicted single consumption amount.

Then, intelligent analysis calculation is carried out aiming at the optimization target of the advertiser.

For example, for an advertising store in different stages of development:

in the first stage, a new store opens and needs to acquire popularity, namely, the exposure, the click rate and the conversion rate are acquired, and at the moment, the optimization weight of the conversion rate/ROI can be weakened to some extent, namely, the part of stores alpha_iThe weight may be higher, β_iSecond, γ_iAnd minimum. Such as alpha_iCan be 0.5, beta_iCan be 0.4, then gamma_iIs 0.1.

In this embodiment, whether one store is a new store, an old store, or a new store may be determined according to the length of the opening time of the store, for example, two time thresholds, that is, a first time threshold and a second time threshold may be preset, where the first time threshold is smaller than the second time threshold, the store with the opening time smaller than the first time threshold is the new store, the store with the opening time between the first time threshold and the second time threshold is the old store, and the store with the opening time larger than the second time threshold is the old store.

In the second stage, after a new store obtains a certain amount of popularity, the quality of the new store can be improved, the promotion of the new store can be partially reduced, and at the moment, an advertiser mainly wants to consume idle resources in the store, namely, the transaction amount is optimized under the condition of ensuring a certain ROI. The amount of the advertisement budget reflects the amount of the currently idle resources. For this part of the advertiser/store/advertisement, the conversion and the deal amount are optimized with emphasis, beta_i、γ_iThe weight may be higher.

In an exemplary embodiment, the ROI may have different definitions in different applications, e.g., ROI ═ sales profit/ad cost 100%, or ROI ═ ad drainage profit/total ad investment. Wherein, the advertisement drainage profit is the total profit-natural single profit; the total advertisement investment is the online investment, the offline investment and the activity investment. For example, the ROI of a certain e-commerce platform on the day of dual 11 is calculated as ROI (total cash profit of orders generated on the day-total profit of average orders on saturday inactive)/(online ad + offline ad + other impressions).

In this embodiment, if the advertising budget investment is not changed, that is, the consumption is not changed, the denominator in (ROI ═ transaction amount/consumption) is not changed, and optimizing the numerator transaction amount optimizes the ROI.

In particular, β can be adjusted by the budget consumption ratio "consumption/advertising budget_i、γ_iThe size of these two parameters. For example, at 70% budget consumption, the beta of the store can be set_i、γ_iRespectively 0.45 and 0.45, and 80% of the budget consumption rate, the beta of the store can be set_i、γ_i0.4 and 0.5 respectively.

In the third stage, when the new store becomes the old store, assuming that the idle resources are consumed,one of the reflective indicators in the advertisement system is that the advertisement budget of the current day/week/month/year (which is also a time range parameter that can be adjusted according to the specific application scenario) has been consumed. For the part of merchants, the transaction amount brought to the merchants by the advertising promotion can be increased under the condition of ensuring that the investment (advertising consumption cost) is not changed, so that the ROI of the merchants is improved. In this case, the emphasis is to optimize the transaction amount, and γ thereof can be set_iHigher weight, beta_iNext, the method is described.

Aiming at the analysis, the historical data of each store can be analyzed, the stage of each store is determined, and the size proportion of different weights is set based on the statistical data in a differentiated mode, so that an advertiser does not need to understand too much and operate a complex advertising system.

Ideally, each store can be divided into a new store, a new store and an old store according to the length of the store opening time, but in actual situations, the stores on the platform may run for a period of time before advertising is put into effect. If the store runs well, the budget consumption ratio is higher even if the advertisement is just put into the store, so that the size of the budget consumption ratio can be used for judging whether the store is a new store, a new and old store or an old store.

The following is the above-mentioned α_i，β_iAnd gamma_iThe detailed calculation method of (1):

first, the CRT was analyzed_i、CVR_iAnd the approximate data distribution of the price of each commodity or service in the store corresponding to the ith advertisement is normalized by a function according with the distribution.

Taking the price of the transaction amount as an example for explanation, assuming that the price in the store is 10-200 units per unit, the whole data is normally distributed, the median is 100, and the simplest normalization way is: normalized trading value is (trading value-100)/(max 200-min 10).

Similarly, CRT_i、CVR_iThe data of (a) conform to a distribution function that can be normalized by the median of the distribution function.

After normalization, combine the rowsSequence formula to find out the alpha of the current reference_i，β_iAnd gamma_iCoefficient of and a_i+β_i+γ_i＝1。

In addition, the reference α is_i，β_iAnd gamma_iThe coefficient can be the coefficient alpha after the previous on-line experiment is finished_i，β_iAnd gamma_iAs a reference coefficient for the current on-line experiment. The present disclosure is not limited thereto.

For optimization goals of varying advertiser granularity, for alpha_i，β_iAnd gamma_iThe adjustment coefficients of the three terms are distributed as a, b and c, then:

1. for a merchant or a store with a budget consumption ratio lower than 50% (the value is a value that can be autonomously set according to a specific situation), mainly optimizing exposure and click, and increasing the weight of a, the weight of a in this embodiment may be relatively calculated by using the budget consumption ratio, for example, when the budget consumption ratio of a certain store is lower than the median of the budget consumption ratios of the same class stores in the same region of the store (for example, in the same city or in the same business circle), then a is calculated as the median of the budget consumption ratios of the same class stores in the same city of the store/the budget consumption ratio of the store. E.g. the store's initial alpha_i，β_iAnd gamma_iThe coefficients are 1/3, and the initial values of a, b and c are 1, then a is alpha initially_i+b*β_i+c*γ_i1 is ═ 1; when the budget consumption ratio of the store is lower than 50%, the increase a is 2, and a is alpha_i+b*β_i+c*γ_i4/3, i.e. not equal to 1, in which case α may be_i，β_iAnd gamma_iAre divided by 4/3, alpha again_i、β_i、γ_iThe sum is again equal to 1. Namely:

α_i＝a*(1/3)/(4/3)＝2*1/4＝1/2；

β_i＝b*(1/3)/(4/3)＝1/4；

γ_i＝c*(1/3)/(4/3)＝1/4。

thus increasing alpha_iThe click weight value of (1).

It should be noted that, in the above embodiment, the calculation a is performed by dividing the median of the budget consumption ratios of the same category in the same city of the store by the budget consumption ratio of the store, but the quantile of the median may be adjusted, for example, to be a 30% quantile.

A Median (also called Median), a term used in statistics, represents a value in a sample, population or probability distribution, which can divide a set of values into two equal parts. For a finite number set, the median can be found by ranking all observations high and low. If there are an even number of observations, the median is usually taken as the average of the two most intermediate values.

The quantile is a variable value at each position of the quantile after all data of the whole are arranged in the order of size. If all data is divided into two equal parts, it is the median; if the quartile is divided into four equal parts, the quartile is obtained; an octant is an octant, etc. The quartile, also known as the quartile point, is the division of the total data into equal four portions, each portion containing 25% of the data, and the value at each quantile point is the quartile.

2. For the merchants or stores with higher budget consumption ratio (e.g. between 50% and 90%, which is an adjustable range), and the merchants or stores with conversion tracking, in this embodiment, for the worst 20% merchants or stores with conversion, conversion is mainly optimized, and the weight of b is increased, and b can be calculated as 20% (which is an adjustable parameter) quantile of exposure conversion rate of the same category in the same city of the store/exposure conversion rate of the store.

The exposure conversion, i.e., the conversion amount, is here the ratio of the conversion after the exposure, and is actually equal to the exposure click rate multiplied by the click rate.

3. Also for merchants with budget consumption ratios that are particularly high (e.g., > 90%, which is an adjustable parameter), the ROI is primarily optimized, weighting c is increased, and raising the ROI further stimulates advertisers to increase advertising budget.

If a is α in the above steps 1 and 2 and 3_i+b*β_i+c*γ_iK, and a, b, and c are limited to a certain valueIn the range of [0.8,1.2 ]](this is also an adjustable range). Finally, three coefficients of the granularity of stores in the ranking are adjusted:

α_i＝a*α_i/k；

β_i＝b*β_i/k；

γ_i＝c*γ_i/k。

it should be noted that, in the above formula, α on the left side of the equation_i、β_iAnd gamma_iIs the adjusted value, α, to the right of the equation_i、β_iAnd gamma_iAre the values before adjustment and a, b and c on the right of the equation are the values after adjustment.

In step S260, the objects are sorted according to the sorting feature value.

In the prior art, the user experience of the C (Consumer, generally referred to as an individual or a home user) end is considered (only the browsing amount/the click amount and the conversion amount are considered), and the weight between the browsing amount and the conversion amount is consistent for all merchants. The ranking method provided by the embodiment of the disclosure starts from the optimization goal of the merchant/advertiser/store, increases the transaction amount/ROI as the feature, and can balance the user experience.

In addition, according to the ranking method provided by the embodiment of the disclosure, different advertisers have different development appeal on click rate, conversion rate, transaction amount/ROI and the like in different development stages, and different weights can be set on three different targets for each advertiser to perform optimization. The method disclosed by the embodiment of the disclosure can intelligently analyze the characteristics and optimization targets of different merchants besides comprehensively considering three factors of click rate, conversion rate and transaction amount, and set the weight of merchant granularity differentiation for the respective different optimization targets of each merchant, and each advertiser can optimize the experience of a B (Business) end by unifying the weights of several optimization targets (click rate, conversion rate and transaction amount) into the same sorting mechanism according to the self development stage. In some embodiments, under the condition that the bids of advertisers are not changed, the possible consumption transaction amount of the users on the advertisements under each PV is estimated in real time, the transaction amount of single promotion is optimized, and the input-output Ratio (ROI) of the advertisers is improved. On the other hand, the method of the embodiment does not need too much participation of the advertiser (such as actively setting the target to be optimized), and can intelligently analyze and determine the target to be optimized of the advertiser.

In this embodiment, pv (page view), which is the page view amount, is usually the main index for measuring a website. The number of web pages viewed is one of the most common indicators for evaluating the website traffic, abbreviated as PV. The pages in the Page Views generally refer to ordinary html web pages, and also include dynamically generated html contents such as php and jsp. One html content request from the browser would be treated as one PV, accumulating into a PV total.

It should be noted that the sorting method shown in fig. 1 and 2 may be applied to a server, where the server may be a system background implemented by a server or a cloud server, and the disclosure is not limited thereto.

It should be noted that the sorting apparatus shown in fig. 3 may be applied to a server, where the server may be a system background implemented by a server or a cloud server, and the disclosure is not limited thereto.

As shown in fig. 3, the sorting apparatus 100 provided in this embodiment may include: an object acquisition module 110, a feature value acquisition module 120, and a ranking module 130.

The object obtaining module 110 may be configured to obtain a plurality of objects to be sorted according to the request information.

The eigenvalue obtaining module 120 may be configured to obtain an ordering eigenvalue corresponding to each object.

The ordering module 130 may be configured to order the plurality of objects according to the ordering attribute.

In an exemplary embodiment, the user attributes include any one or more of taste preferences, environmental preferences, price sensitivity, brand preferences of the current user.

In an exemplary embodiment, the eigenvalue acquisition module 120 may include a historical data acquisition sub-module and an eigenvalue operator module. The historical data acquisition submodule can be used for acquiring historical exposure, historical click rate and historical single amount of each object. The eigenvalue operator module can be used for obtaining the click eigenvalue, the conversion eigenvalue and the transaction eigenvalue of each object according to the request information, the historical exposure, the historical click rate and the historical order placing rate.

In an exemplary embodiment, the feature value operator module may include an exposure click rate calculation unit and a click feature value calculation unit. Wherein the exposed click rate calculation unit may be configured to obtain a historical exposed click rate of each object according to the historical exposure and the historical click rate. The click feature value calculation unit may be configured to obtain a click feature value of each object according to an object attribute of each object, the historical exposure click rate, and the request information.

In an exemplary embodiment, the feature value operator module may include a click down rate calculation unit, an exposure conversion calculation unit, and a conversion feature value calculation unit. The click order rate calculation unit may be configured to obtain a historical click order rate of each object according to the historical click amount and the historical order amount. The exposure conversion rate calculation unit may be configured to obtain a historical exposure conversion rate of each object according to the historical exposure click rate and the historical click rate. The conversion feature value calculation unit may be configured to obtain a conversion feature value of each object according to an object attribute of each object, the historical exposure conversion rate, and the request information.

In an exemplary embodiment, the feature value operator module may include a transaction amount predictor, a normalization unit, and a transaction feature value calculation unit. The transaction amount pre-estimation unit may be configured to obtain a predicted transaction amount of each object according to the object attribute of each object and the request information. The normalization unit may be configured to normalize the predicted transaction amount for each subject. The trading feature value calculation unit may be configured to obtain the trading feature value according to the normalized predicted trading amount and the conversion feature value of each object.

In an exemplary embodiment, the sorting apparatus may further include a weight obtaining module. The weight obtaining module may be configured to obtain a click weight, a conversion weight, and a transaction weight of each object according to a current state of each object.

In an exemplary embodiment, the weight obtaining module may include a first weight setting sub-module and/or a second weight setting sub-module and/or a third weight setting sub-module. The first weight setting submodule can be used for setting the clicking weight of the object to be greater than the conversion weight and the conversion weight to be greater than the transaction weight when the object is in the first state. The second weight setting submodule can be used for setting the conversion weight of the object to be more than or equal to the transaction weight and the transaction weight to be more than the click weight when the object is in the second state. The third weight setting sub-module may be configured to set a transaction weight of the object to be greater than a conversion weight and a conversion weight to be greater than a click weight when the object is in a third state. And the sum of the click weight, the conversion weight and the transaction weight of each object is a preset constant.

In an exemplary embodiment, the first weight setting sub-module may include a first weight calculation unit. The first weight calculation unit may be configured to increase the click weight of the object according to the consumption budget ratio of the object when the consumption budget ratio of the object is within a first preset range.

In an exemplary embodiment, the second weight setting sub-module may include a second weight calculation unit. The second weight calculation unit may be configured to increase the conversion weight of the object according to the conversion characteristic value of the object when the consumption budget ratio of the object is within a second preset range.

In an exemplary embodiment, the third weight setting sub-module may include a third weight calculation unit. The third weight calculation unit may be configured to increase the transaction weight of the object according to the input-output ratio of the object when the consumption budget ratio of the object is within a third preset range.

The ranking device provided by the embodiment of the disclosure starts from the optimization target of a merchant/advertiser/store, increases the transaction amount/ROI as a characteristic, and can balance the user experience.

In addition, with the ranking device provided by the embodiment of the disclosure, different advertisers have different development appeal on click rate, conversion rate, transaction amount/ROI and the like in different development stages, and different weights can be set for each advertiser on the three different targets for optimization.

It should be noted that, for specific implementation of the modules of the sorting apparatus in the foregoing embodiment of the invention, reference may be made to the contents of the sorting method in the foregoing embodiment of the invention shown in fig. 1 and 2, and details are not described here again.

According to another exemplary embodiment of the present disclosure, there is also provided an electronic device, which may include a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the program, when executed by the processor, implements the method steps in the inventive embodiments illustrated in fig. 1 or fig. 2 described above.

Referring now to FIG. 4, shown is a schematic diagram of an electronic device 400 suitable for use in implementing embodiments of the present application. The electronic device shown in fig. 4 is only an example, and should not bring any limitation to the functions and the scope of use of the embodiments of the present application.

As shown in fig. 4, electronic device 400 includes a processor 401 that may perform various suitable actions and processes in accordance with programs stored in a memory 403. In particular, according to an embodiment of the present disclosure, the processes described above with reference to the flowcharts may be implemented as computer software programs. For example, embodiments of the present disclosure include a computer program product comprising a computer program carried on a computer readable medium, the computer program containing program code for performing the method illustrated in the flow chart, which when executed by a processor 401, performs the above-described functions defined in the system of the present application. The processor 401, the memory 403, and the communication interface 402 are connected to each other by a bus.

The flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of terminals, servers, methods and computer program products according to various embodiments of the present application. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams or flowchart illustration, and combinations of blocks in the block diagrams or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

The modules described in the embodiments of the present application may be implemented by software or hardware. The modules described may also be provided in a processor.

As another aspect, the present disclosure also provides a computer-readable medium, which may be contained in the apparatus described in the above embodiments; or may be separate and not incorporated into the device. The computer readable medium carries one or more programs which, when executed by a device, cause the device to comprise: acquiring a plurality of objects to be sequenced according to the request information; acquiring a sorting characteristic value corresponding to each object; and sorting the plurality of objects according to the sorting feature values; the sorting characteristic value corresponding to each object is obtained based on the click characteristic value and the corresponding click weight of each object, the conversion characteristic value and the corresponding conversion weight, and the transaction characteristic value and the corresponding transaction weight.

Exemplary embodiments of the present disclosure are specifically illustrated and described above. It is to be understood that the present disclosure is not limited to the precise arrangements, instrumentalities, or instrumentalities described herein; on the contrary, the disclosure is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims.

Claims

1. A method of sorting, comprising:

the method comprises the steps that a plurality of objects to be sequenced are obtained according to request information, the request information comprises search information input by a current user, combination information between the current user and each object and user attributes of the current user, the search information is search keywords input by the current user, the combination information is distance, direction and traffic condition information between the current user and each object and matching degree between the search keywords input by the current user and each object, and the user attributes comprise personalized information of the current user;

acquiring historical exposure, historical click rate and historical lower unit rate of each object;

acquiring the historical exposure click rate of each object according to the historical exposure and the historical click rate; obtaining a click characteristic value of each object by adopting a first machine model according to the object attribute of each object, the historical exposure click rate and the request information, wherein the object attribute comprises the class, the business district and the geographical position information of each object;

obtaining the historical click order rate of each object according to the historical click amount and the historical order rate; obtaining the historical exposure conversion rate of each object according to the historical exposure click rate and the historical click order rate; obtaining a conversion characteristic value of each object by adopting a second machine model according to the object attribute of each object, the historical exposure conversion rate and the request information;

obtaining a predicted transaction amount of each object according to the object attribute of each object and the request information; normalizing the predicted transaction amount for each subject; obtaining a transaction characteristic value by adopting a third machine model according to the normalized predicted transaction amount and the normalized conversion characteristic value of each object;

acquiring the click weight, the conversion weight and the transaction weight of each object according to the current state of each object;

acquiring a sorting characteristic value corresponding to each object based on the click characteristic value and the corresponding click weight of each object, the conversion characteristic value and the corresponding conversion weight, the transaction characteristic value and the corresponding transaction weight; and

sorting the plurality of objects according to the sorting feature values;

the step of obtaining the click weight, the conversion weight and the transaction weight of each object according to the current state of each object comprises the following steps:

when the consumption budget ratio of the object is in a first preset range, judging that the current state of the object is in a first state, and increasing the click weight of the object according to the consumption budget ratio of the object so as to enable the click weight of the object to be larger than the conversion weight and the transaction weight;

when the consumption budget ratio of the object is in a second preset range, judging that the current state of the object is in a second state, and increasing the conversion weight of the object according to the conversion characteristic value of the object so as to enable the conversion weight of the object to be larger than or equal to the transaction weight and the click weight;

and when the consumption budget ratio of the object is within a third preset range, judging that the current state of the object is in a third state, and increasing the transaction weight of the object according to the input-output ratio of the object so as to enable the transaction weight of the object to be larger than the conversion weight and the click weight.

2. The method of claim 1, wherein obtaining a click weight, a conversion weight, and a transaction weight for each object based on the current state of each object further comprises:

when the object is in the first state, setting the conversion weight of the object to be greater than the transaction weight; and/or

When the object is in the second state, setting the transaction weight of the object to be greater than the click weight; and/or

When the object is in the third state, setting the conversion weight of the object to be greater than the click weight;

and the sum of the click weight, the conversion weight and the transaction weight of each object is a preset constant.

3. The method according to any one of claims 1 to 2, wherein the request information includes search information input by a current user and/or combination information between the current user and each object and/or user attributes of the current user.

4. A sequencing apparatus, comprising:

the object acquisition module is used for acquiring a plurality of objects to be sequenced according to request information, wherein the request information comprises search information input by a current user, combination information between the current user and each object and user attributes of the current user, the search information is search keywords input by the current user, the combination information is distance, direction and traffic condition information between the current user and each object and matching degree between the search keywords input by the current user and each object, and the user attributes comprise personalized information of the current user;

the weight acquisition module is used for acquiring the click weight, the conversion weight and the transaction weight of each object according to the current state of each object;

the characteristic value acquisition module is used for acquiring a sequencing characteristic value corresponding to each object; and

the sorting module is used for sorting the plurality of objects according to the sorting characteristic values;

the sorting characteristic value corresponding to each object is obtained based on the click characteristic value and the corresponding click weight of each object, the conversion characteristic value and the corresponding conversion weight, and the transaction characteristic value and the corresponding transaction weight;

wherein, the characteristic value acquisition module comprises:

the historical data acquisition submodule is used for acquiring the historical exposure, the historical click rate and the historical lower unit amount of each object;

the exposure click rate calculation unit is used for acquiring the historical exposure click rate of each object according to the historical exposure and the historical click rate; the click characteristic value calculation unit is used for obtaining the click characteristic value of each object by adopting a first machine model according to the object attribute of each object, the historical exposure click rate and the request information, wherein the object attribute comprises the category, the business district and the geographical position information of each object;

the click order rate calculation unit is used for acquiring the historical click order rate of each object according to the historical click rate and the historical order rate; the exposure conversion rate calculation unit is used for acquiring the historical exposure conversion rate of each object according to the historical exposure click rate and the historical click rate; a conversion characteristic value calculation unit, configured to obtain a conversion characteristic value of each object by using a second machine model according to an object attribute of each object, the historical exposure conversion rate, and the request information;

the transaction amount pre-estimating unit is used for obtaining the predicted transaction amount of each object according to the object attribute of each object and the request information; a normalization unit for normalizing the predicted transaction amount of each object; the transaction characteristic value calculating unit is used for acquiring the transaction characteristic value by adopting a third machine model according to the normalized predicted transaction amount and the normalized conversion characteristic value of each object;

wherein the weight obtaining module comprises:

the first weight setting submodule comprises a first weight calculating unit, a second weight calculating unit and a processing unit, wherein the first weight calculating unit is used for judging that the current state of the object is in the first state when the consumption budget ratio of the object is in the first preset range, and increasing the click weight of the object according to the consumption budget ratio of the object so as to enable the click weight of the object to be larger than the conversion weight and the transaction weight;

the second weight setting submodule comprises a second weight calculating unit and a second weight setting submodule, wherein the second weight calculating unit is used for judging that the current state of the object is in a second state when the consumption budget ratio of the object is in a second preset range, and increasing the conversion weight of the object according to the conversion characteristic value of the object so as to enable the conversion weight of the object to be more than or equal to the transaction weight and the click weight;

and the third weight setting submodule comprises a third weight calculating unit and a third weight setting submodule, wherein the third weight calculating unit is used for judging that the current state of the object is in a third state when the consumption budget ratio of the object is in a third preset range, and increasing the transaction weight of the object according to the input-output ratio of the object so as to enable the transaction weight of the object to be greater than the conversion weight and the click weight.

5. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the program realizes the method steps of any of claims 1-3 when executed by the processor.

6. A computer-readable medium, on which a computer program is stored which, when being executed by a processor, carries out the method steps of any one of claims 1 to 3.